[KAMI911]

Summary

Fixes #886 (every keyboard layout switch freezes the compositor for up to ~1s).

• reload_keybindings() unconditionally ungrabbed and regrabbed every per-window keybinding (Alt+F4, Super+arrows, etc.) on every window, on every keyboard layout group switch. Each grab/ungrab is a synchronous XIGrabKeycode/XIUngrabKeycode round trip, so this blocks Cinnamon's main loop (it's the X11 compositor) for O(windows × per-window bindings) round trips per switch — @ashvedovcds measured 290ms–960ms with 32–149 open windows in X11: every keyboard layout switch freezes the compositor for up to ~1 s (reload_keybindings regrabs per-window keybindings on every window) #886.
• The per-window bindings in question (F-keys, arrows, Space) resolve to the same keycodes in every layout, so the full regrab does no useful work in the common case. It's only needed when a layout switch actually changes a resolved combo's keycodes or modifier mask (e.g. non-Latin layouts, which fall back to a secondary US layout for some keysyms).
• This implements "option 1" from X11: every keyboard layout switch freezes the compositor for up to ~1 s (reload_keybindings regrabs per-window keybindings on every window) #886's Expected behavior section: compare old vs. new resolved combos and only pay for the regrab when something changed, rather than unconditionally skipping/doing it (their "option 2").

Implementation

• Added per_window_combos_will_change(): a read-only probe that resolves each META_KEY_BINDING_PER_WINDOW binding's combo under the incoming layout (after reload_modmap(), before reload_combos() overwrites anything) and compares it against the binding's current resolved combo via a new resolved_key_combo_equal() helper, without mutating any binding state.
• Split ungrab_key_bindings()/grab_key_bindings() (previously root window + full per-window loop) into the cheap root-only operation plus new ungrab_per_window_key_bindings()/grab_per_window_key_bindings(). reload_keybindings() now only calls the per-window variants when the probe finds a real difference. The META_PREF_KEYBINDINGS preference-change path (user edits a shortcut) keeps calling both unconditionally, since that's rare and not the perf-sensitive path.
• The always-cheap root-window grab/ungrab is unchanged and still runs unconditionally on every reload, so non-per-window bindings (and non-Latin-layout per-window bindings) are unaffected.

Test plan

• Reproduce X11: every keyboard layout switch freezes the compositor for up to ~1 s (reload_keybindings regrabs per-window keybindings on every window) #886's measurement script (pingsw.py) with ~100+ open windows on an X11 session with 2+ keyboard layouts; confirm the post-switch Cinnamon D-Bus freeze drops to near-baseline when per-window bindings use layout-independent keys (arrows/F-keys/Space)
• Switch to a non-Latin layout (e.g. ru) with a per-window binding whose keysym needs the secondary US layout fallback, and confirm the per-window regrab still happens and the binding still works after the switch
• Verify editing a keybinding in Cinnamon Settings (META_PREF_KEYBINDINGS path) still correctly re-grabs both root and per-window bindings